[asterisk-dev] Speech Recognition

[Stephan A. Edelman]

----- Kevin P. Fleming wrote:
> When we first designed the Asterisk/speech interaction, we were
planning
> on doing exactly that. However, after going through LumenVox's
training
> and speaking to others, we realized that speech recognition apps (at
least
> successful ones) are not really designed to just let people say 'one',
> 'two', 'three', etc., so we focused on letting the user build
applications
> that directly manipulate grammars and react to the responses.

I'm not sure I follow this reasoning. To use VoiceMailMain() as an
example, it is quite trivial to update the necessary voice prompts to
say "press 6 or say 'next' for next message", etc. and use a grammar
that equates the word 'next' with the result 6. It still means that we
need a transparent way to do voice recognition. If this facility was
available, in my opinion, this would significantly enhance almost any
application that accepts DTMF input.

> In spite of that, though, it would be fairly straightforward to create
an > application like SpeechDTMF() that would just sit on the channel
(using 
> the chanspy infrastructure), feed audio to the speech engine and then
send
> DTMF frames in when the engine recognized the appropriate phrases.
Patches
> are welcome :-)

I hadn't thought about the chanspy facilities, thanks for the heads up!
I'll get started on it. 

Regards,

Stephan.